FEDERATED ENSEMBLE LEARNING WITH METAHEURISTIC OPTIMIZATION FOR PRIVACY-AWARE SOFTWARE DEFECT PREDICTION

Abstract

The software defect prediction is an important issue in software quality as it helps to detect the defective software modules at early stages of software development. Cor rect prediction models decrease the cost of maintenance, testing and total project risk. Conventional software defect prediction methods, however, tend to be centralized and use standard machine learning methods, which could have issues with class imbalance, data privacy, heterogeneous data distribution, and lack of model generalization. These limitations have led to a need for advanced and scalable prediction frameworks that can perform effectively in real-world environments. This thesis introduces a thorough research to apply federated learning and hybrid op timization methods to software defect prediction. The main emphasis of the work is on the three research contributions. The first contribution involves developing a FedProx based federated learning framework with the aim of enhancing the model’s generalization performance in a decentralized setup, whilst simultaneously managing Non-IID data dis tributions amongst the clients. The second contribution presents a MOON-based model contrastive federated learning framework to further reinforce the model consistency at local sites while boosting the convergence performance in distributed scenarios. The fourth one introduces an integrated ensemble learning–PSO–GWO optimization frame work for better feature selection and hyperparameter tuning and enhance the classification accuracy for defect prediction. Overall, this thesis highlights the potential of combining federated learning, meta heuristic optimization, and ensemble learning to build efficient and scalable software defect prediction systems. The findings of this work can contribute toward the development of intelligent and privacy-aware software engineering solutions capable of handling complex and distributed software development environments.